Even though you may never purchase a share in Anthropic, there’s a high possibility that money in your portfolio is already riding on the success of businesses like it.
Nvidia (NVDA) sells the chips powering artificial intelligence. Amazon (AMZN), Alphabet (GOOGL), Microsoft (MSFT), and other technology giants are spending enormous sums building the cloud and data-center infrastructure that AI developers need.
Near the opposite end of that chain are Anthropic and OpenAI. They must transform all of that processing power into something that consumers will pay for.
Investor Steve Eisman, well-known for his involvement in the subprime mortgage trade detailed in “The Big Short,” interprets Anthropic’s most recent figures as a warning.
Eisman said on CNBC that Anthropic’s sequential revenue growth slowed from 147% in the second quarter to 45% in the third quarter. He also said revenue generated relative to compute and infrastructure spending deteriorated.
That does not mean Anthropic is shrinking. Third-quarter revenue reached $16.8 billion, up from $11.6 billion in the second quarter, Anthropic said in a filing.
However, Eisman’s worries extend beyond Anthropic’s first public offering.
Investors may ultimately have to reevaluate how much infrastructure the whole AI boom needs if the firms using massive quantities of processing power are unable to continue converting that expenditure into quickly increasing income.
Eisman sees a change in Anthropic’s growth story
For almost any corporation, Anthropic’s first half 2026 results would seem exceptional.
Revenue reached $16.3 billion during the six months ended June 30, up from $1.2 billion a year earlier, according to Anthropic’s IPO filing.
It also had a very good second quarter, bringing in $11.6 billion. Revenue was somewhat more than double its compute expenditure, with compute and infrastructure costs totaling more than $5.6 billion.
Anthropic highlighted that improving efficiency in its prospectus, noting that second-quarter revenue jumped 1,368% from a year earlier, while compute and infrastructure expenses increased 300%.
Eisman is examining the subsequent events.
Earlier in October, Anthropic informed potential investors that its third-quarter sales reached $16.8 billion, a 45% sequential rise.
That growth is still really quick.
According to Eisman, the contrast with the prior quarter, when sales rose by almost 147% from the first quarter, is the issue.
Additionally, he determined that Anthropic made around $1.70 in revenue for every dollar spent on computing and infrastructure in the third quarter, compared to roughly $2.08 in the second.
Eisman said Anthropic’s ability to turn compute into revenue decreased throughout the quarter as a result of those numbers. “The whole chain depends upon the health of Anthropic and OpenAI.”
Eisman attributes the shift in part to the termination of “token maxing.”
When users give prompts and get responses, AI models analyze the text fragments known as tokens. Businesses use more tokens and often spend more money when they deploy AI models more extensively.
Eisman uses the term “token maxing” to characterize the time when companies urged employees to use AI extensively, with no regard for the expense.
He contends that businesses are now becoming more mindful of what those AI tokens really do. The second half of 2026 may therefore be a more significant test of sustained demand than the previous spending boom.
Why Anthropic matters to Nvidia and the rest of the AI trade
This is when an obscure efficiency ratio becomes considerably more significant for the typical investor.
A vast amount of computational infrastructure is required for Anthropic.
Compute is its biggest operational expenditure, according to its prospectus, and it has significant obligations. Crucially, regardless of Anthropic’s final income or usage, many of those obligations often have to be fulfilled.
Reuters estimates Anthropic has committed more than $600 billion to computing and infrastructure partners over the coming decade.
This expenditure supports a much wider ecology. Anthropic says Claude operates across Amazon Web Services Trainium chips, Google Tensor Processing Units (TPUs), Nvidia graphics processing units, and AMD (AMD) GPUs.
Additionally, it distributes Claude via Google Cloud, Microsoft Azure, and AWS.
This implies that Anthropic serves as a source of demand, a collaborator, and a consumer for some of the largest firms driving the AI investment surge.
Therefore, whether Claude is a decent chatbot is not the only issue for someone who buys Nvidia stock. Another factor is whether businesses like Anthropic can generate enough revenue from AI services to support their ongoing use of additional processors, data centers, power, and cloud capacity.
The same reasoning applies to Amazon and Alphabet, two major suppliers of Anthropic’s infrastructure and distribution as well as investors in the company.
Eisman’s argument may be significant to a typical tech investor because of this.
The expenditure cycle becomes simpler to explain if Anthropic can continue to generate much more income from each additional dollar of computer capability.
The math gets uncomfortable if infrastructure needs continue to rise and GDP continues to decline.
A “Big Short” trader spots a warning inside Anthropic’s growth.OLIVER CONTRERAS / Getty Images
Anthropic has plenty of evidence against the bearish case
Treating a quarter of reduced growth as evidence that the AI boom is ending is another serious flaw.
Anthropic continues to develop at a very rapid rate.
Its third-quarter revenue of $16.8 billion was more than 3.5 times the $4.7 billion it generated in the first quarter, according to Reuters. The company also told prospective investors that it was profitable on an adjusted basis during the quarter.
Additionally, its clientele offers a compelling refutation of the notion that Claude demand is declining.
Anthropic had more than 11,900 customers generating at least $100,000 of trailing-12-month revenue as of Sept. 30. A year earlier, it had more than 950.
Approximately 250 clients made $10 million or more, while more than 1,900 generated at least $1 million.
Additionally, Anthropic reported a net revenue retention of almost 900%, indicating a significant increase in the spending of current clients throughout the studied time. Over half of Fortune 500 organizations generated at least $100,000 in trailing-12-month sales.
These figures don’t indicate that a company is abruptly losing clients.
Additionally, Anthropic advises investors to anticipate a gradual slowdown and quarterly fluctuations in its revenue growth rate.
That is typical math for a business growing significantly.
Growing $4.7 billion of quarterly revenue by 147% is very different from growing $16.8 billion at the same rate.
When CNBC questioned whether Eisman’s theory implied a sharp decline in the whole market, he likewise refrained from forecasting an impending collapse in technology stocks.
“I’m not willing to make that bet just yet,” he said. That difference is important.
No one claims that the AI trade has already failed. Investors now have a new figure to watch.
Anthropic’s next numbers could test the entire AI spending story
With Anthropic’s IPO, public-market investors will get regular financial results from one of the firms closest to the core of the generative-AI expenditure surge, something they haven’t received before.
Because of this, its future revenue growth and compute efficiency are significant factors that go well beyond the final price of ANT shares.
Anthropic’s quarterly growth of 147% is not necessary for an investment in Nvidia. However, a future in which AI engineers always want more processing power and can afford to pay for it helps Nvidia and other infrastructure suppliers.
Anthropic’s own prospectus explains that link in detail.
According to the corporation, in order to keep access to financing, it must show that it can make profitable investments, transform computing investment into improved models, and then translate those advancements into consumer acceptance. On the other hand, when revenue declines, its computational duties do not instantly decrease.
Investors should monitor that cycle.
Increased income is a result of increased client utilization. Greater computing is supported by higher income. Demand for semiconductors, cloud infrastructure, and data centers rises as computing increases.
What worries Eisman is what would happen if that circle began to slow down at the first step.
Anthropic’s 45% quarterly sales increase seems significant.
The warning comes from the direction of the numbers: Growth slowed sharply from the previous quarter, while Eisman’s measure of compute efficiency also fell.
A trend cannot be established in a single quarter. Still, Anthropic’s next quarters may reveal far more to investors than whether its initial public offering (IPO) was priced appropriately, given the hundreds of billions of dollars invested in AI infrastructure and the fact that some of the world’s biggest IT businesses are betting on that continued demand.
They might help determine whether the economics behind the AI boom are keeping pace with the money being invested in it.
Related: Nvidia’s $3 billion Meta bet reveals a 30-fold shift